IOSR Journal of Polymer and Textile Engineering (IOSR-JPTE)

e-ISSN: 2348-019X, p-ISSN: 2348-0181, Volume 2, Issue 1 (Jan - Feb. 2015), PP 34-39
www.iosrjournals.org

Scope for Artificial Neural Network in Textiles

C. S. Sentthil Kumar 1, Dr. M. Dhinakaran 2
1

(Department of Fashion Technology , Kumaraguru College of Technology, India)

(Department of Textile Technology , Kumaraguru College of Technology, India)

Abstract: The artificial neural network (ANN) is increasingly used as a powerful tool for many real world
problems. ANN has proved its usefulness for resolving many problems in textiles such as prediction of yarn
properties, analysis of fabric defects, process optimization etc. The power of neural networks lies in their ability
to represent complex relationships and learn them directly from the data being modeled. The prediction of
properties or performance of a process in advance is required to minimize the setup cost and time. The ability to
predict these properties accurately has become a challenge due to highly non-linear and interactive behaviour
of textile materials. This paper presents basics of ANN and its applications in different textile domains.
Keywords: Back propagation, Cotton grading, Fabric engineering, Image processing, Multilayer feed forward
network

I.

Introduction

The artificial neural network (ANN) is an information processing paradigm that is inspired by the way
biological nervous system works, such as brain processes an information. ANN uses numeric and associative
processing to mimic models of biological system. It resembles the human brain in two respects: i.e. knowledge
is acquired by the network from its environment through a learning process and interneuron connection
strengths, known as synaptic weights, are used to store the acquired knowledge [1].
Today, ANN is being applied to increasing number of real world problems of considerable complexity.
The advantage of ANN lies in their resilience against distortion in the input data and their capability of learning.
They are often good at solving problems which are too complex for conventional technologies e.g. problems
that do not have a conventional algorithmic solution [2].
1.1 Definition
A set of processing units when assembled in a closely interconnected network, offers a rich structures
exhibiting some features of the biological neural network. Such a structure is called as artificial neural network
(ANN). The ANN is a massively parallel distributed processor made of single processing units, which has a
natural propensity for storing experiential knowledge and make it available for use[1].
The procedure used to perform the learning process is called as learning algorithm, the function of
which is to modify the synaptic weights of the network in an orderly fashion to attain a desired design objective.
In practice, the ANN cannot provide the solution by working individually, rather it needs to be integrated into a
consistent system engineering approach[3].
1.2 History
In 1943, Warren McCulloch and Walter Pits proposed a model of computing element, called as
McCulloch Pits neuron. In 1949, Donald Hebb proposed a learning scheme for pre-synaptic and post-synaptic
values of variables. In 1958, Rosenblatt proposed the perceptron model, which can adjust weights by perceptron
learning law. In 1960, Widraw proposed Adaline model for computing element. In 1985, Akley, Hinton and
Sejnowski proposed feedback neural network with hidden units. For many years, neural networks have been
used in various areas of engineering and economics in order to describe the relationship between the parameters
that cannot be determined analytically. They were used to simulate learning from examples, detecting patterns,
associative memorizing and recalling information[2].
1.3 Classes of Neural Network
There are three classes of neural networks: viz. Single layer feedforward networks, Multilayer
feedforward networks and Recurrent networks as shown in Fig. 1, 2 and 3 respectively.
1.3.1 Single layer feed forward network
The neurons are organized in the form of layers. In a simplest form of a layered network, an input layer
of source nodes projects onto an outer layer of neurons. The network is strictly feedforward and hence called as
single layer feedforward network[1].
DOI: 10.9790/019X-0213439

www.iosrjournals.org

34 | Page

Scope for Artificial Neural Network in Textiles

Fig.1 Single layer NN

Fig. 2 Multilayer NN

Fig. 3 Recurrent NN

1.3.2 Multilayer feed forward network

In multilayer network, there is presence of one or more hidden layers whose nodes are called as hidden
neurons. Their function is to intervene between external input and network output in some useful manner. By
adding more hidden layers, the network is enabled to extract higher order statistics. The source nodes in input
layer supplies respective element of activation pattern, which constitutes input signals applied to neurons in first
hidden layer. The output signals of the first hidden layer are used as input to the second hidden layer and so on.
The set of output signals of the neurons in the final output layer of network constitutes the overall response of
the network to the activation pattern supplied by the source nodes in the first input layer.
1.3.3 Recurrent network
The recurrent neural network is distinct in the sense that, it has at least one feedback loop. A recurrent
network may consist of a single layer of neurons, with each neuron feeding its output signals back to inputs of
all other neurons. The recurrent network may or may not have hidden neurons. The presence of feedback loops
has a profound impact on learning capability of the network and its performance[1].
1.4 Learning of a network
A network can be learnt with or without training. The objective of learning is to mold the mapping
surface according to desired response. Once the network is structured for a particular application, that network is
ready to be trained. To start with, the initial weights are chosen randomly. There are two approaches to learning:
supervised and unsupervised.
1.4.1 Supervised learning
In this system, both input and output data are provided to network. Then it processes the inputs and
compares the resulting outputs against the desired outputs. Errors are propagated backwards through the system,
causing system to adjust the weights. The same set of data is processed many times, as the weights are refined.
When the system has been fully trained, no further learning is required and the weights can be frozen. A vast
bulk of networks utilizes supervised learning.
1.4.2 Unsupervised learning
This type of network is provided only with inputs. The system itself decides which features to be used to group
the input data. It is also called as Adaptive learning. This learning technique is not commercially practiced[4].

II.

Applications Of Ann In Textiles

2.1 Fibres
An Artificial neural network is used by S. Mukhopadhay and Q. Siddiquee [4] for the prediction of
copolymer components. The free radical copolymerization of acrylamide with quaternary ammonium cationic
co-monomer diethylaminoethyl acrylate (DMAEA) has been investigated in inverse emulsion. The copolymer
composition is determined from residual monomer concentration. An ANN is used to predict the copolymer
composition. The results shown that, ANN can predict copolymer composition very correctly, as a function of
reaction conditions and conversions.

DOI: 10.9790/019X-0213439

www.iosrjournals.org

35 | Page

Scope for Artificial Neural Network in Textiles

Fig. 4 Intelligent fibre classification system

Fig. 4 shows the intelligent fibre classification system. Here, ANN is used to classify two kinds of
animal fibres; Merino and Mohair. The model extracting six scale parameters with image processing and other
using an unsupervised neural network to extract features automatically, which are determined in accordance
with the complexity of scale structure and the accuracy of the model. It can achieve higher accuracy by
increasing number of samples for learning, since accuracy of ANN largely depends on the parameters selected
[5]. Cotton trades in India assess the commercial value of cotton on the basis of "grade" assigned by hand and
eye methods by professional cotton classers. Now, ANN can be used to develop models to predict the cotton
grading. By this way, cotton grading can be made more reliable. Research is going on in this direction to replace
the existing subjective grading with the objective one. To be precise, the properties of cotton that are measured
by using ANN are colour of cotton, trash present and ginning preparation. The blend out of two different kinds
of fibres, the exact proportion in which the mixing has to be done (to obtain the required properties of the yarn)
can be predicted using ANN [6].
2.2 Spinning
Y.C. Zeng et al [7] have analyzed modeling methods for predicting the tensile properties of air jet spun
yarns. Numerical simulation provides a useful insight into a relationship between yarn tenacity and parameters.
The neural network model predicts yarn tenacity using input parameters. The predicted and experimental values
agreed well, indicating that the neural network is the excellent method for predictors. They also analyzed the
effect of input parameters on yarn tenacity with neural network model.
M. C. Ramesh et al [8] predicted the tensile properties of yarn from fibre properties. The feasibility of
yarn engineering is demonstrated in Fig. 5, by developing a yarn to fibre "reverse" model using ANN. This
approach is entirely different from the prevailing forward models, which predict the properties of final yarn by
using the fibre properties as inputs. The cost minimization of cotton fibre was ensured by using classical linear
programming approach in combination with ANN. The engineered yarns demonstrated good agreement with the
targeted yarn properties. Spinning consistency index (SCI) was chosen for this purpose, as the regression
equation of this index comprises most of cotton properties.

Fig. 5 ANN model for yarn engineering

Luo. Cheng and David Adams [9] used an ANN for prediction of relationship between fibre properties
and yarn strength. A three layer feedforward network along with a correlation regression is used. They used
USDA cotton fibre and processing test results for this investigation. The fibre properties like upper half mean
length, length uniformity, short fibre content, strength; fineness and maturity ratio are measured using HVI. All
cotton samples were spun into 22s ring spun yarn. Yarns were tested for skein test, elongation, appearance and
DOI: 10.9790/019X-0213439

www.iosrjournals.org

36 | Page

Scope for Artificial Neural Network in Textiles

imperfection. The CSP was used as a prediction target. When the neural networks were trained by a training set
CSP values could be predicted from the neural networks in the prediction mode. Then the predicted CSP were
correlated with the actual CSP and a good agreement is found to be existing between fibre properties and yarn
strength.
Chung Feng Jeffrey et al [10] studied the relationship between process variables and molecular
structure. The thermo-mechanical properties of PET yarns were determined using ANN. Although the internal
structure of ANN is incomprehensive with respect to mechanisms of modeled relations, it was accurate way to
make data and relations between data sets easily accessible. Das A. [11] used a multilayer feed forward network
with back propagation algorithm to determine the optimized machine settings of a draw texturing machine, with
regard to desired yarn properties. The properties of yarn like tenacity and elongation are predicted by using
regression analysis and the process parameters used are draw ratio, D/Y ratio, speed and primary heater
temperature.
Traditionally, the quality grades of false twist yarn packages are classified by human inspection, but the
result may be affected by personal and subjective factors. Shih-Hsuan Chu et al [12] used the image processing
technology to extract the defects in yarn packages. They used neural networks to classify the quality grades of
the yarn packages. From the experimental results, they obtained the classifying rate of about 90%.
2.3 Knitting
S. Ertugrul and N Ucar [13] studied the prediction of bursting strength of knitted fabrics. Bursting
strength of cotton plain knitted fabrics is predicted before manufacturing by an intelligent technique of neural
network and neuro fuzzy approaches, by achieving a significant savings in cost and time required for trial and
error. Among many parameters that affect fabric bursting strength; the fabric weight, yarn breaking strength and
elongation are the input elements for this prediction. Prediction of total hand value of knits is reported to be
successful by Park [14]. A fuzzy neural network provides an effective tool for prediction of total hand value of
outerwear knit fabrics. A fuzzy neural network is developed to predict and display the drope images of garments
made of different fabrics and styles. The new approach is used to develop a prototype drope prediction system,
to predict drope of women's dress styles made from different fabrics.
2.4 Weaving
Rocco Furferi et al [15] presented a work to provide a predictive model of a coating process for
forecasting the final characteristics of a coated fabric, based on the process parameters. The ANN is trained and
validated using a wide experimental database. The work is proved to be useful for technicians in selecting the
optimal processing parameters for obtaining the desired coated fabric properties. The developed system requires
technicians to set the values of process parameters and to stimulate the ANN based model to forecast the quality
of coated fabric.
An image processing system is used by Atiqul Islam et al as a tool for dynamic inspection of fabrics;
the inspection sample is a piece of plain white fabric. The four defects are holes, oil stains, warp lacking and
weft lacking. The image treatment employs a high resolution linear scan digital camera. Fabric images are
acquired first, and then the images are transferred to a computer for analysis. Finally, the data is adopted as
input for neural network which is obtained from readings after treating the images. In this system, there are three
feed forward networks; an input layer, one hidden layer and an outer layer. As it has the ability to cope with the
non-linear regression property, this method can reinforce the effects of image identification [16].
Automatic recognition of woven fabric patterns presented by Boong S. J. et al [17]. The neural network
and image processing technology are introduced for classifying woven fabric patterns. An auto correlation
function is used to determine one weave repeat of the fabric. The reflected fabric image is captured by CCD
camera and digitized by computer system. The learning vector quantization's algorithm as a learning rule of the
artificial neural network enables recognition of woven fabric types more effectively. The results demonstrated
that, the three fundamental weaves can be classified accurately and structural parameters such as yarn spacing,
its variance and the ratio of warp to weft spacing can also be obtained. The system consists of a PC and CCD
(charged coupled device) camera equipped with an attachable zoom lens. The fabric image is captured by a
lightening device which is mounted at the end of the camera and zooms in on the image, which is also displayed
on the coloured monitor. The image is then digitised by the AC to DC convertor and processed using techniques
such as thresholding and auto-correlation to analyse the fabric pattern. The original image is converted into a
grey level image to improve computer processing time and speed and the histograms are obtained.
Jasper W. J. [18] used the back propagation algorithm to study the missing ends and picks, oily fabrics
and broken fabric, all often found as a defect in fabric. The correct selection of parameters for input layer plays
a great role in recognition rate. Once a defect occurs in a fabric, its periodicity is changed so that corresponding
intensities at specific position of spectrum obviously changed. These intensions can act as characteristics
parameters and can be substituted in ANN for learning. In a study [19] by Tsai L. S. and Hu M. C., the
DOI: 10.9790/019X-0213439

www.iosrjournals.org

37 | Page

Scope for Artificial Neural Network in Textiles

efficiency and accuracy of a method of detecting fabric defects that have been classified into different categories
by neural network. Four fabric defects most likely found during weaving were learned by network. The method
used for processing image is co-occurrence based method, by which six parameters are obtained. The results
shows that, the fabric defects inspected by means of image reorganization in accordance with ANN agrees with
initial expectations.
The Uster fabric scan system [20] uses ANN in evaluating wrinkled fabrics with image analysis.
Wrinkles in cloth usually develop with deformation during storage. It is not easy even for trained observers
judge the wrinkles. Here an objective method is being proposed. The visual information stimulates their senses
and makes them to judge the grade of wrinkle. Since visual evaluation is complex and bias, we cannot use linear
evaluation system

Fig. 6 Uster Fabriscan

For auto inspection of wrinkles in fabric. For this purpose neural networks are widely used. Since
networks are non-linear, the training rule is regularized as non-linear optimization problem.
Input parameters such as angular second monument, contrast, correlation, entropy and fractal
dimension are obtained using image analysis and are fed to the neural network algorithm. The mean sensory
value presenting the grade of wrinkled fabric as output is obtained with desired parameters. The parameters to
be perceived such as density, size, form and contrast can be inspected using multi-input and multi-output
concept of neural network. A specific algorithm named "Kalman's Filter Algorithm" is used for this purpose and
results are referred to be very much feasible for evaluating wrinkled fabrics.
The relationship between structure and properties of the fabric is complex and inherently non-linear; to
create a predictive model one must resolve these complexities. The optimization of input parameters required to
produce a fabric as per end use is called as fabric engineering [21]. Fabrics can be engineered either by weaving,
knitting or bonding neural networks are successfully implemented in all three to optimize the input parameters.
The yarn properties and machine settings are used as inputs to predict the fabric properties as outputs. The
construction of neural network for fabric engineering is shown in Fig.7.

Fig. 7: Neural network for fabric engineering

DOI: 10.9790/019X-0213439

www.iosrjournals.org

38 | Page

Scope for Artificial Neural Network in Textiles

In prediction of human psychological perceptions of clothing sensory comfort ANN is widely used. The
objective is to know the predictability of the clothing sensory comfort from psychological perceptions by using
feed forward back propagation network in ANN system. In order to achieve the objective, a series of wear trials
are conducted in which 10 sensory perceptions (clammy, clingy, damp, sticky, heavy, prickly, scratchy, fit,
breathable and thermal) and overall clothing comfort is rated. Good agreement between predicted and actual
clothing comfort perceptions proved that, the neural network is effective technique for modeling the
psychological perceptions of clothing sensory comfort. It is also found that, the predicted comfort score from the
model with hidden neurons and the linear output neuron has a better fit with the actual comfort score than other
models with different combinations of hidden and output neurons. Compared with statistical modeling
techniques, the neural network is a fast and flexible predictive tool with a self learning ability for clothing
comfort perceptions [22].

III.

Conclusion

Ultimately, an ANN is one of the hopes available to textile industry to integrate elements such as
production, properties, quality, cost, and statistical process control and process optimization. ANN is field of
computer based technology that can simulate characteristics of human interference for industrial benefits. ANN
aims at the requirements of the industry like least downtime, highly competitiveness and reliable estimates of
properties and process parameters. The function of ANN is not constant but can be changed dynamically.

References
[1].
[2].
[3].
[4].
[5].
[6].
[7].
[8].
[9].
[10].
[11].
[12].
[13].
[14].
[15].
[16].
[17].
[18].
[19].
[20].
[21].
[22].

Haykin Simon, "Neural Networks- A comprehensive Foundation", 2 nd edition, Prentice Hall Of India Pvt. Ltd., (2006), pg 2, 21.
B. Yegnanarayana, "Artificial neural networks", 5 th edition, Prentice hall of India Pvt. Ltd., Sep. (2007), pg 24,118.
M. Ananda Rao and J. Srinivas, "Neural networks- algorithms and applications", 2 nd edition, Narosa Publishing House, (2005), pg
17-31.
S. Mukhopadhay and Q. Siddiquae, "Artificial neural networks and their use in textile technology", Asian Textile Journal, Vol. 56,
March, (2003), pg 72-77.
She F. H., Kong L. X., Nahavandi S. and Kouzani A. Z., "Intelligent animal fibre classification with artificial neural network",
Textile Research Journal, Vol. 72, (7), (2002), 594-600.
Xu B., Su J., Dale D. S. and Watson M. D., "Cotton colour grading with neuron networks", Textile Research Journal, Vol. 70,
(2000), 430-436.
Y. C. Zeng, K. T. Wang and C. W. Yu, "Predicting the tensile properties of air-jet spun yarns", Textile Research Journal, Vol. 74,
(6), Aug. (2004), 689-694.
M. C. Ramesh, R. Rajamanickam and S. Jayaraman, "The prediction of yarn tensile properties using by using artificial neural
networks", Journal of Textile Institute, Vol.86, (3), (1995), 459-469.
Luo Cheng and David L. Adams, "Yarn strength prediction using neural networks- Part 1: Fibre properties and yarn strength
prediction", Textile Research Journal, Vol. 65, (9), (1995), 495-500.
Chung Feng Jeffrey Kuo, Kun Iuan Hsiao and Yi Shiuan Wu, "Using neural network theory to predict the properties of melt spun
fibres", Textile Research Journal, Vol.74, (9), (2004), 840-843.
Das A., "Network application in texturing process and parameters", "Texturing viabilities- Proceedings", The Textile Association
India- Mumbai unit, 95-116, (2005).
Shih-Hsun Chu, Hung Ming Chen and Jyh Yeow Chen, " Appearance analysis of false twist textured yarn packages using image
processing and neural network technology", Textile Research Journal, Vol.71, (4), April (2001), 313-317.
S. Ertugrul and N. Ucar, "Predicting bursting strength of plain knitted fabrics using intelligent techniques", Textile Research
Journal, Vol.70, (10), (2000), 845-851.
Park, S. W. Hwang, Y. G. and Kang B. C., "Applying fuzzy logic and neural networks to total hand evaluation of knitted fabrics",
Textile Research Journal, Vol. 70, (8), (2000), 675-681.
Rocco Furferi, Lapo Governi and Yary Volpe, "Modelling and simulation of an innovative fabric coating process using artificial
neural networks", Textile Research Journal, Vol. 82, (12), July (2012),1282-1294.
Atiqul Islam, Shamim Akhter and Tumnun E, Mursalin, "Automated textile defect recognition system using computer vision and
artificial neural network", Proceedings of world academy of Science, Engineering and Technology, No.13, May, (2006), 1-6.
Boong Soo Jeon and Ji Hyun Bae, "Automatic recognition of woven fabric patterns by artificial neural network", Textile Research
Journal, Vol.73, (7), (2003), 645-650.
Jasper W. J. and Potlapalli H., "Image analysis of miss picks in woven fabrics", Textile Research Journal, Vol. 65, (11), (1995),
683-692.
Tsai L. S. and Hu M. C., "Automatic inspection of fabric defects using ANN", Textile Research Journal, Vol. 66, (7), (1996), 474482.
R. Meier, Uster Fabriscan: The intelligent fabric inspection, http://www.kotonline.com/ana_baslikar/ uster.asp
Fan K., Wang T., Cheng T., Jou C. and Kou T., "Fabric classification based on recognition using a neural network and
dimensionality reduction", Textile Research Journal, Vol. 68, (3), (1998), 179-185.
Wong A. S. W., Li Y. and Yeung P. K. W., "Neural network predictions of Human Psychological perceptions of clothing and
sensory comfort", Textile Research Journal, Vol. 73, (1), (2003), 31-37.

DOI: 10.9790/019X-0213439

www.iosrjournals.org

39 | Page